协同过滤推荐算法的改进与研究  

Improvement and Research of Collaborative Filtering Recommendation Algorithm

作　　者：孙华艳 李业丽 字云飞 韩旭 SUN Hua-yan;LI Ye-li;ZI Yun-fei;HAN Xu(School of Information Engineering,Beijing Institute of Graphic Communication,Beijing 102600,China)

机构地区：[1]北京印刷学院信息工程学院,北京102600

出　　处：《计算机技术与发展》2018年第10期44-48,共5页Computer Technology and Development

基　　金：国家自然科学基金(11603004);北京市科技创新服务能力协调创新项目(PXM2016_014223_000025);北京市科技创新服务能力建设科研水平提高定额项目(2017-04190117010);北京市自然科学基金项目(1173010)

摘　　要：针对协同过滤算法中数据稀疏性严重影响推荐精准度的问题,提出了基于因果聚类和模糊相似关系的协同过滤推荐算法。基于因果聚类的协同过滤推荐算法通过对用户—项目的评分、收藏、特征值等建立矩阵来计算它们之间的相似度,矩阵的建立是基于项目、用户特征值具有因果关联性和相似性;基于模糊相似关系的协同过滤推荐算法通过项目、用户的加权特征法构建三角模糊数来计算相似度,并根据模糊相似矩阵及最佳阈值法来得到连通图截集,从而实现精准、个性化推荐。实验结果表明,这两种算法都能有效提高相似度计算精度,解决数据稀疏性问题,从而提高推荐的精准度和个性化。且随着最近邻数量的增加,基于模糊相似关系的协同过滤推荐算法的精准度略高于基于因果聚类的协同过滤推荐算法。Aiming at the problem that data sparsity seriously affects the accuracy of recommendation in collaborative filtering algorithms,we propose a collaborative filtering recommendation algorithm based on causal clustering and fuzzy similarity relation. Collaborative filtering recommendation algorithm based on causal clustering builds matrices to compute their similarity through the user-project scoring,collection,eigenvalues and so on. The establishment of matrix is based on the causality and similarity between the project and the user’seigenvalue. To achieve accurate,personalized recommendations,the collaborative filtering recommendation algorithm based on fuzzy similarity relation computes the similarity through constructing triangular fuzzy number by the weighted feature method of project and user,and gets connected graph truncation set according to the fuzzy similarity matrix and the best threshold method. Experiment shows thatboth algorithms can effectively improve the accuracy of similarity calculation,solving the problem of sparse data and raising the precisionand personalization of the recommendation. As the number of nearest neighbors increases,the collaborative filtering recommendation algorithm based on fuzzy similarity relation is slightly more accurate than the collaborative filtering recommendation algorithm based oncausal clustering.

关 键 词：推荐系统 协同过滤 因果聚类 模糊相似关系 最大树 阈值 加权 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]